TUSET. 1. K. S. Thc rcfining or silicol1 and rcrrosilicon. INFACON fl. ProceediIlR.~ of Ilu! 6111 ImemarioJlal Fcrrotll/oys Co"~re.\·;},, CapeTO\llIl. Volume I. Johanllcsburg. SAIMM. 1992. pp. 193-199.

The Refining of Silicon and Ferrosilicon

J. K. s. TUSETSINTEF Metallurgy. Trondheilll. Norway

The types and origins of impurities in ferrosilion and silicon metal, and the optionsin the refining of the liquid metal are reviewed.

The paper, which deals only with the chemistry of oxidative refining, focuses oncarbon and oxygen solubilities in liquid iron-silicon alloys and on distribution equi­libria for calcium and aluminium between the metal and ternary CaD-Si02-AI2D]slags. The impact of these elements on dissolved oxygen in the met..1 phase is alsodiscussed.

Distribution data are presented i.n diagrams showing isoconcentration lines for cal­cium and aluminium in the liquid range of the CaD-Si02-AbO] system at 1550 °e.Reaction paths and kinetic obstacles in the refining process are explained with refer­ence to these diagrams, and the precautions to be taken to avoid the formation ofox.ide inclusions are mentioned.

IntroductionWith an annual production 400 kt of Ferrosilicon and 130 ktof silicon metal in 1989. the power-intensive Norwegiansilicon industry produced about three-fifths and one-third ofWestern Europe's total output of these products respective­ly. To maintain this industry in a leading position as a reli­able supplier of quality products, substantial research-and­development work on liquid- and solid-state refining hasbeen carried out over the past few dccades by the industryitself and also at research institutes. This work has not beenterminated. however, and more work is still needed in orderto meet the requirements of demanding customers, some ofwhom are asking for specifications that sometimes seem tobe beyond the theoretical limitations.

Before the industry can be in a position to respond ration­ally to such requests, it needs to know, firstly, the theoreti­cal limitations and, sccondly, how to approach them in areproducible way in actual production.

This paper, which deals with the oxidative retining ofliquid metal only, focuses on the first point: What are thetheoretical limitations in oxidative refining, and what typeof supplementary data is needed to quantify them?

Origin and Control of hnpuritiesIn any metal-production process, quality control starts withthe selection of raw matcrials. This statement should in par­ticular be emphasized to producers of a reactive metal suchas silicon and its alloys, where the selectivity of the processitself is nearly non-existent, and where the possibilities forbulk metal refining are very limited.

Because of the nature of the metal itself, the carbothennicprocess for the production of silicon and ferrosilicon is inprinciple def1l1ed as a slag-free process. This implies that,apart from elements forming gaseous components andvolatile metals or metal oxides that are lost with the off­gases, the elements entering with the charge materials areexpected to be reduced and tapped as constituents of the

THE REFINING OF SILICON AND FERROSILICON

product metal. A variety of metallic impurities arc thereforepresent, and those usually recorded in commercial productsare listed in Table I.

However, aluminium and calcium, being the most abun­dant impurities in the charge, do not" behave as simply asstated above. Aluminium appears as an impurity in bothquartz (or quartzites) and the reductants used, the latterbeing responsible For the input of 60 to 70 per cent of thetotal aluminium and 90 to 95 per cent of the total calcium ina normal charge used for the production of standard fer­rosilicon I. The oxides of these elements present in thereduction materials are expected to be reduced fairly easily,whereas the alumina, which is diluted in the silica phase, isless available for reduction and tends to accumulate as aviscous silicate melt in the furnace. When this melt is con­tacted with metal sufficiently high in calcium, an exchangereaction takes place that increases the CaO content and thefluidity of the oxide phase. Occasionally, therefore, someslag is tapped together with the metal. On average, slag tapMping in standard ferrosilicon production represents an outletfor 30 to 40 per cent of the total aluminium and 50 per centof the total calcium in a charge ' .

Titanium, the third major impurity metal in standard fer­rosilicon and silicon metal, originates from both the SiO"source and the reduction materials used, the latter being themain source. Titanium is practically absent in tapped slag.however, and nearly 100 per ccnt reports to the metal phase.The same applies to metallic impurities from groups V 1OVII of the Periodic Table, which in the case of fcn"osiliconentcr with the iron source, in the form of scrap iron, iron­orc pellets, or mill scale. These impurities are not usuallyrecorded in standard ferrosilicon. bUl have specified maxi­mum levels in high-purity products, as indicatcd ill Table I.

In the reduction furnace, the metal forms in the presenceof silicon carbide at temperatures in the range 1800 to 20000c. The carbon content of tapped metal therefore reflects thesolubility of carbon in ferrosilicon at these temperatures.

193

TABLE IIMPURITY LEVELS TN SOME TYPICAL FERROSILICON AND SILICON-METAL PRODUCTS. % IlY MASS

>\< Accordmg to specificatIons by the producer (ILS). Maximum levels specified on 22 elements lor the Fe66S1 qualnyr Mainly present as SiC inclusions (Figure I)

Fcrrosilicon S iIicon metal

Elcment Standard Relined High-plJrilY High-purity Standard RefinedFe75Si Fc75Si Fe75Si* Fe66Si* MGSi MGSi

Si 74-79 65 -67 97 - 99 > 98,5

AI < 2,0 0,05 -0,50 < 0,06 < 0,020 < 0,6 0,1-0,5

Cu 0,2 - 0,8 0,05 -0,10 <0,02 < 0,020 0,2 0,Q3 - 0,Q7

Ti 0,07-0,10 0,10 - 0,20 <0,04 <0,016 0,05 0.05

Ct 0,05 -0.15 0,02 <0.02 < 0.012 0.Q3 0.Q2

Fe (Balance) (Balance) (Balance) 0,3 - 0,5 0,3 -0.5

Mn <0,130 0.Q3 0.03V. Cr, Ni, Cu 0,01 - 0,Q3 (WI - 0,03 0,0 I - 0,(13

Co, Mo. Zr < 0,003 < 0,005 < 0,005

S < 0,00 I - 0,005

P 0.015 - 0,Q3 0.0110 0.005 0.005

B < 0,0005 0,005 0,004-

The values reported for the solubility of carbon in liquidsilicon vary. The data of Scace and Slack' form the basis oflhe solubility curve shown in Figure 1. This figure includesI.In enlarged portion of the Si-C phase diagram at the melt­ing point of silicon as given by Nozaki el (fe' in order toshow the eutectic nature of the system. For the ternary sys­tem Fe-Si-C, reliable data on carbon solubi lily exist onlyup to the stability range of SiC, which appears4 al around23 per cent silicon and 0,4 per cent carbon at 1600 'c. Theshape of the solubility curve from this point up to the Si-Cbinary, along which the activity product of carbon and sili­con has to take a constant value, is not known.

RepOited data4 from two experiments with an alloy con­taining 45 per cent silicon, together with preliminary resultsof ongoing work at the SINTEF laboratories, indicate that

the solubilities given in Figure I are fairly representative ofcommercial ferrosilicon. The carbon analysed in theseproducts is therefore mainly present as primary particles ofprecipitated and suspended silicon carbide that are trappedduring the solidification process.

Oxygen, which is as inevitable as carbon in the product,is not listed in Table L It is, however. an element of grow­ing concern lo the producers of high-quality silicon and fer­rosilicon.

The solubility of oxygen in liquid Fc-Si alloys ill equilib­rium with solid silica at 1600 and 1650 'C has been studiedfairly extensively by NovokhaLskiy5 and Shevtsov6-M andtheir co-workers. Their results are in fairly good agreement,and are presented in Figure 2, which shows the data ofShcvtsov at 1650 °C. It can be seen that there is a minimum

1800

Slrl) 0,08

1700 s/c+S/(/)()0

~ 0,06"]ii 1600 fQ) SI(I) (0)%a.E ~ 0.02

SIC~SI(/)

Q) <;SI~SI(1)

0,04I-

r~1500 ~.""

0 0.001 0,""Wolght'l' C

0,02

Jl-

1/ --.,

,/

V/

~/1400

o

S/+Sll

51

0,Q1

S/+S/C

0,02 0,03

c; % by rna-55·0,04 0,05 o 20 40 60

(Si)%80

FIGURE 1. Solubility of carbon in liquid silicon. derivcd from the dlltll orSCllCC llnd Slack2. including the culeclil; portion of the Si-e system as

givcn by Nozaki el al. 3FIGURE 2. Oxygen solubility in Fe-Si melts at 1650 "C

(after Shevtsovll)

194 INFACON 6

Oxygen, % by mass

FIGURE 3. 'llC system Si-O with solubility curves :IS given byNovokhalskys. The dash-dolled curve represellts the liquid solubility as

given by HiraI:''}

in oxygen concentration at low silicon contents. as can berecognized from the known deoxidation equilibriu of liquidsteel, whereas a maximum in oxygen solubility appears ataround the composition of Fe75Si. The solubility value at100 per cent silicon :.IS given in this diagram is consistentwith data for tilt system Si-Si02 taken from the work ofNovokhatskiy el al. 5. Their dmil were llsed in the construc­tion of Figure 3. which shows (he silicon-rich part of thissystem. The uncertainty in these data is illustmled by thesolubility curve. which was calculated from the reccnt workor Hirata and l-Ioshikawa9.

It is worth noting that the diilgr;Jm shows the feature of ;Jcutectic system. One would expect this binary eutectic toextend towards a ternary eutectic in the system Si-Fe-O.The location of this eutectic. which is not known. isassumed to be fairly close to the Fe-Si eutectic appearing at58,2 per cenl silicon and 1207 dc. The level of oxygen atthis ternary eutectic will probably be of the order of 50p.p.m. Therefore. in commercial alloys containing morethan 50 to 80 p.p.m. of oxygen. the excess oxygen would beexpected to be present as primary oxide inclusions.

Options in the RefIning of Liquid MetalApart [rom sophisticated methods like electrolytic andvacumm refining. both of which are considered to be ofinterest in connection with the production of solar-grade sil­icon. there are only two options of prdctical interest in Ihebulk liquid-met~l1 relining of silicon metal and ferrosilicon:

methods based on the oxidation and slagging of im­purities.methods based on chlorination and the removal of im­purities as volatile chlorides.

Both types of mel hod are effeclive for the same types ofreactive elements such as the alkali and alkaline-earth met­als and aluminium. The chlorinat"ion process. being themore effective of the two for these elements, appears to alsohave a certain purification effect on other elements such aschromium, manganese, and copper. The process is. how­ever. of less industriill interest owing to the environmentaland material problem:) associated with the use of chlorinclind the emission or corrosive metal chlorides. The siliconindustry therefore uses only methods that are based on theoxidation and slagging of impurities.

where the underlining indicates elements that are dissolvedin the metal phase.

A dissolved metal impurity, Me, present at the interfacemay now react as follows:

II J

121

14]

131

1/2 SiD, (dissolved ill slag) = 1/2 Si + Q

..J'2'. SiOl (slag) + Me = f Si + 1 Me.•O,. (slag)~ _x x .

with the equilibrium constant

KMr.Si = (asiP'I2.\·. (aMe.p.) /Ix

aMI'· (a:,'iOZ) yllx

~ MJ:.. + Q ~ 1, MexO\· (dissolved ill ,,·Ill}:).

If both reactions are at local equilibrium at the interrm:e.with the same ac[ivity of dissolved oxygen (ao), thcn thisquantity can be eliminated if equations r11 and [2J are com­bined. Multiplicaton of the resulting expression by the fac­tor y/x yields the following equiltion:

Distribution Equilibria in Oxidative ReliningNo matter how the oxygen is introduced. a boundary-layerfilm of oxide forms at the metal surface as a result of thereaction between the oxygen and the most abundant ele­ment. which is silicon. The reaction defin.ing the oxygenpotential in the metal at this intcrface is. then

The oxygen needed for the refining reactions in oxidativerclining can be introduced

;'IS i.I gas in the fOfm of oxygen or air :.Il the metal surface,or by gas-blowing through <I hlllce. a nozzle, or plug inthe bottom of a refining vesselin the form of an oxidizing agent. such as Fe20] in thecase of ferrosilicon. or Sial in the case of silicon metalor by slag treatment combined with mixing.

Gas-blowing is usually combined with the addition ofsome kind of slag-forming compounds that may also act asoxidizing agents. The addition is usually carried out byinjection when a lance is used for gas-blowing, or by directcharging to a top slag in the case of the nozzle-blowingtechnique. where intimate mixing associated with air-blow­ing can be obtained. The various sources of oxygen listedare therefore effective in a real process, but their relativeimportance may differ from onc type of operation to theother. The way the oxygen is introduced does, in fact.strongly affect the heat billelnce and the kinetics of the relin­ing process, but has in principle no intluence on the chem­cial equilibria controlling the process.

This is equivalent to stelling that the oxidation of i.l dis­solved impurity is coupled with the oxidation of silicon,since both reactions appear to be under the control of theS<llllC rate-limiting slep, here assumed 10 be the rale ul whichoxygcn is supplied 10 the metal-slag interface. If this is true,it is also evident that the oxidation of the various dissolvedelements is mutually coupled through reactions similar tothose shown in equation l3]. At low impurity levels, howcv­er. the transport of the impurity elements from the bulk ofthe metal to the slag-metal interface may very well takeover as 11 rate-controlling step, and the resulting distributionwill deviate from the equilibrium values discussed below.

Table 11 lists the set of distribution reactions obtainedwhen Me is substituted for d.l. and en in equation l31. The

20 30 40 50 60

Uq.oxld

S/(I/'/Iq,lHhH 'i1700

510,- .

1<110

1416.96

'800

1700 51(1)

~l' '800~

S/(IJ • Chd.'omliJ.iii"0-

"00E{".

u ..SI(IJ. rr/d)<mJtl'

"00 -51 51 • rtfdymIM

1300

0.05 0.'

THE REFINING OF SILICON AND FERROSILICON 195

TABLE IIDISTRIBUTION EQUILlBRIA 1N THE OXIDATIVE REFINING OF LIQUID FERROSILICON AND SILICON METAL

Calculated k values

Fe75Si as solvent Si metal as solvent

Distributing equilibria and corresponding t1G, kJexpressions for impurity concentrations

k/" = (as/)ynx MMc [Q + ~ ] ( V'" 100 M~lckl*=~K M Si MF.· K MSj

1550°C 1600°C 1550°C 1600°C 1550°C 1600°C

[3.1] dl + ~ (SiD,) =i ill: + ~ (AI,O,)-107,307 -106,742

I (a ) In[%Al] = k- Alz01

3/4¥AI (aSiO]) 6,09 10" 7,63 10" 8,06 10" 1,01 10.2

[3.2] C!l + i (510,) = f ill: + (CaO)-145,017 -140.306

I a[%Ca] =k- CoO

YCu (aSiO?) In 7,89 10" 1,38 10" 9,94 10" 1,74 10"

* Inserted for aSi - 0,82 In Fc7SS1 and aSi - 1 In 51 metal. MMe - mole masst The calculations were based on data from JANAp lO, except for the free energy of formation of cristobalitc, where the data of Rein and Chipman I I were

used

When these values are used as shown in Table II, two setsof distribution constants are obtained. One set of valuesrefers to Fe75Si, and the other to silicon metal as the sol­vent for impurities. In both cases, the constants given arecorrect only for impurities present at infinite dilution. Thisdepends on the silicon activity values used (0,82 and 1,0 forFe75Si and silicon metal respectivelyl2) and the correctionfactors used for the change in concentration from mole frac­tions to percentage by mass.

With this limitation in mind, the given constants can be usedin the calculation of the equilibrium concentration of each ofthe two elements listed provided that their activity coefficients

standard free energies of these reactions at 1550 and1660 °C are also listed. All the data used in these calcula­tions are from JANAplO, except for the standard free energyof cristobalite, which was derived from the equation givenby Rein and Chipman ":

8G',,;,,, = - 226,500 + 47,S T (kcal/mol). [5j

This expression yields values that deviate slightly fromthe JANAF values at 1550 and 1600 'C, but it was chosento obtain consistency with the slag-activity data of Rein andChipman II. Furthermore, it has to be pointed out that thet1G values given for reaction no. 3.2 in Table II are not tileproper standard values, since liquid calcium was chosen asthe reference state in these calculations, not gas at a pres­sure of 1 atmosphere.

The equilibrium constants are calculated from the well­known expression

Isoconcentration lines for calcium and aluminium, drawnon the basis of both experimental and calculated data at1550 °C, are shown in Figures 4 and 5 respectively. A dia­gram similar to Figure 5, which shows good agreement, hasbeen published by Ageev and his co-workers I4 , who studiedthe distribution of aluminium between Fe75Si alloys andsynthetic slags at 1600 °G.

(YMc) in the metal solutions, as well as the impurity oxideactivity and tbe SiOl activity of the slag phase are known.

These constants, as well as similar expressions for equi­libria involvi.ng the other impurities appearing in standardproducts, can also be used in the calculation of the expectedactivities of the oxides formed from these impurities whenthe Si02 activity is equal to I, which represents the mostoxidizing conditions that can be realized in this system. Theuse of estimated and conservatlve y-values for the metalimpurities leads to the conclusion that, apart from the sili­con itself, aluminium and calcium are the only metalsamong those present that will contribute significantly to thefonnation of slag during the oxidation process. A relevantslag is therefore the ternary system CaO-A1203-Si02.

When the accepted activity data for this system were usedas given in the comprehensive work of Rein andChlpman II, together with experimental data on the equili­hrium distrihution of calcium and aluminium betweenFe75Si alloys and synthetic slags located within the liquidregion of the system at 1500, 1550, and 1600 'C, good cor­relation between the experimental and the calculated valueswas obtained when the following values were inserted inthe expressions given l3 in Table II:

[7]YCa = 2,1' 10.3 alldYAi = 0,45.

[6](8G")K= exp . RT

196 INFACON6

FIGURE 5. lsoconcenlrntion lines for AI in Fc75Si alloys in equilibriumwith CaO-Si02~AJ20rslags at 1550 °C, in percentages by mass

(aftcr Tusct13)

[ 12]

[II]

[%Calsi = 1,26 [0,006]' 2,6 = 0,Q2.

....Ft!Si~= 2,1' 10-3/8' 10-4 = 2,6.rCa

The use of this value as the correction factor for an alumi­na-free silica-saturated slag, yields the following expressionfor the equilibrium concentration of calcium:

This value agrees fairly well with the experimental valuesobtained at 1550 0c.

A correction factor that is applicable for aluminium inliquid silicon is unfortunately not available at present, butwork on this is in progress.

Oxygen in Refined AUoysNo reliable infonnation is currently available on dissolvedoxygen in real silicon alloys containing calcium, aluminium,and carbon. However, this should not prevent speculation onthis topic.

Calcium, aluminium, carbon, and silicon are known to act asdeoxidizing agents in liquid steel, despite the fact that they alllower the activity of dissolved oxygen in steel, as reflected bythe negative values of their interaction coefficients, i.e. E& < O.The addition of a deoxidizer should, owing to this effect,increase the oxygen level provided ti,e oxygen potential iskept constant. The opposite is experienced, however. Theexplanation is, of course, that it is the deoxidizer itself thatnow defines the oxygen potential of the system, and this takesplace almost right from the beginning of the addition. Thisphenomenon is clearly reflected in the oxygen-solubilitycurve for the Fe-5i alloys shown in Figure 2.

When silicon or a silicon-rich ferroalloy is the solvent foroxygen, the situation is somewhat different. In that case, theoxygen potential is defined by the slability and activity of lheoxide of the solvent itself, and it will even remain constant,independent of the level of calcium and aluminium, as long assolid silica forms as a stable phase. As the interaction coeffi­cients of these elements on dissolved oxygen are expected tobe negative, although less negative than in steel, higher levelsof dissolved oxygen would be expected in metal in equilibri­um with silica-saturated calcia-aJumina slags than in the caJ­cium- and aJuminium-free systems. However, the actual vaJ­ues have yet to be determined experimentally.

For a metal containing higher levels of calcium and alumin­ium in equilibrium with unsaturated slags of decreased silicaactivity, an overall decrease in the concentration of dissolvedoxygen migbt be expected, since it is believed that the oppos­ing effect, caused by the increase in the content of deoxidizer,is more than counterbalanced by the decrease in the oxygenpotential of the system.

For calcium in silicon, the valueSi -yCa=yc,=8'1O-4 [10]

can be used. This value was derived from the work ofSchUnnann el al. 15, and leads lo the relationship

'Y g,Si :J.l!1iS' = 1,26 [%Ca]F,Si 5i [8]

Y~a YCa

k2[%Ca]si = [%Ca]F,si-

k"

U,I~...,."

",SU",'Ill

1.1.)

,.'"'~~~

FIGURE 4. lsoconccntmtion lines for Ci1 in Fc75Si alloys in equilibriumwith CaO-SiOrA 120 r slags at 1550 0c, in perccntages by mass

(after Tuset13)

When the data in these diagrams are applied to siliconmetal in equilibrium with the same slags, the impurity con­centrations have to be corrected relative to the given valuesfor ferrosilicon as follows, with k values taken fromTable II:

The y-ratios appearing in these expressions are correctionfactors for the change in the activity coefficients of calcium:md aluminium respectively during the transition from iron­containing to iron-free alloys. The y-ratios are both expect­ed to be greater than unity, owing to the stronger interactionbetween the iron and the silicon than that between the ironancl the dissolved impurities.

Equilibrium Reaction Paths and Their KineticInterpretations

In the refining of silicon, the density of the slags formed isnearly the same as that of the metal. A silica-rich slag of theternary type considered will float on the silicon-metal bath,whereas the more basic and alumina-rich slags tend to sink 13.

THE REFINING OF SILICON AND FERROSILICON 197

diU,

FIGURE 6. Isoviscosi!y lines (poisc) in thc caO-SiO~-AI:PJ systcmat I500°C (ancr El1ioll el al.l~)

This problem does not occur in the refining or ferrosilicon.The viscosity of the slag is important in the rel"illillg of bothsilicon and ferrosilicon; its variation with composition at1500 cC is shown by the isoviscosity lines in Figure 6.

Consider now the refining of ferrosilicon containing 1.5per cent aluminium and 0.5 per cent calcium being blownwith oxygen or air in a ladle al 1550 0c. II is i.tssumed that aboundary slag layer forms at the metal-gas interface of thegas bubbles being created deep in the ladle. and that thisslag reaches equilibrium with the bulk metal before thebubbles burst at the surface. leaving their slag contentbehind as a top-layer slag. The composition of the slagformed in the initial stage of the refining process is indicat­ed by the letter A in Figure 7. This slag, which has nearlythe same composition as those reported for tapped slagsfrom ferrosilicon furnaces I. has a calcium-to-aluminiumratio larger than that of thc metal. Consequcntly. the metal

FIGURE 7. Isoconcentralioll linc:" for c.a (broken lillcs) anu AI(solid lincs) in Fe75Si at 1550 °e as rcproduccd from pigures 5 and 6.

Path A-B I represents the change in thc composilion Ill' Ille product slagwhen an 111loy eonlaining 1,5 pcr cClll '1luminiulll and 0.5 per I:elll calcilllll

reacts 10 equilibriunl wilh O;l;ygCI1. provided lhe lOp slag is in;lctive.Path A-B~ reprcsents thc samc data for an activc top slaf:. i.e. equilibrium

has been established throughout (after TUSCI 1.,)

198

becomes depleted in calcium.If the top-layer slag does not paJlicipate in the reaction, the

resulting slag changes its composition from A towards B r inFigure 7 as the metal changes its composition from I,S to 1,1per cent aluminium and 0,5 to 0.05 per cent c;,tlcium.

If the top-layer slag also participates in exchange re­actions with the metal, the resulting equilibrium slag com­position is shifted tow,uds point 8 2 in Figure 7. and theconcentrations of aluminium and calcium in the metal willchange to 0,7 per cent and 0,04 per cent respectively.

Along these paths from A to 8 there is a large increase inslag viscosity. Furthermore, for the removal of aluminiumto continue beyond the above limits, a highly viscous andpartly solidified alumina-silicate slag has to form. Onewould therefore expect that the efficiency of the oxidationprocess will decrease gradually as the calcium concentra­tion of the metal decreases. and that the aluminium contentwill tend to level off at a concentration much higher thanpredicted for a silica-saturated slag.

This agrees with observations made during industriallrialsand bench-scale tests. The process is running with nearly 100per cent oxygen utilization in the initial stage when calciumis present ,Hid a liquid slag is formed, but becomes inefficientas soon as the calcium is consumed and a crusty. nearly dryoxidation product of silicon starts to for111. This usuaLlytakes place at around 0.05 to 0.06 per cent calciulll in siliconmetal, and at 0,03 to 0,04 per cent calcium in fcrrosiliconl 6.

Calcium deticiency at the site of oxidation will not onlyinf"luence the oxygen yield as described, but will also have adramatic effect on the oxygcn in the product mctal, owingto the formation of crusty silica films that do not easily dis­solve or scparate from the liquid metal. Such tiny particles.being difficult 10 identify undcr the microscope, will, ofcourse, result in oxygen levels far in excess of the liquidsolubility values.

If oxygen lancing is used and the degree of mixing is lim­ited, the flux should preferably be injected with the oxygen.This is, however. not necessary in an air- or oxygen-blownprocess using i.I bottom plug, as described for the Tinjectprocess t7 , where the top slag appears to participi.tte through­out the reaction owing to intimatc mixing. In that case, adirect addition of nux to the top slag is sufficient.

The carbon removal thaI takes place in oxidative reliningis not considered here, since such refining is not based onchemical reactions but is a separation process based on dif­ferences in densities and interfacial energies.

From reported measurements of contact angles, it is clearthat liquid silicon wets silicon carbide better thun quartz orfused silicn 1x. This does not mean, however. that siliconcarbide is welled better by silicon metal than by silicateslags. On the contrary, all the observations made during themelting of silicon or ferrosilicon in graphite or silicon car­bide crucibles in the presence of slags SUppOI1 the conclusionthat the slag wets silicon carbide better than it docs the metal.Slags are always found enveloping the metal in these cru­cibles, and particles of silicon carbide are always seen con­centrated in the slag phase. The question of relative wettingis therefore not critical, at least in the case of acidic slagswith low surface energies, but it may be for more basic slags.

111e extent to which suspcnded particles of silicon carbidecan be effectively removed in an oxidative refining processtherefore dcpcnds primarily on the tluidity of the slag andlhe degree or intcrmixing of mctal and slag, as well as onthe conditions under which the phases separate.

INFACON 6

AcknowledgmentThe author gratefully acknowledges the financial supportreceived from the Research Association of the NorwegianFerroalloy Producers (FFF) and the Royal NorwegianCouncil for Scientific and Industrial Research (NTNF).

ReferencesI. Oterkjaer. H. (1976). A study of the AI- and Ca-balance

in industrial production of Fe75Si. Dr thesis (inNorwegian). NTH. 123 pp.

2. Scace, R.I.• and Slack, GA (1959). Solubility of carbon insilicon and germanium. J Cllelll. PI,y.f.. 30. pp. 1551-1555.

3. Nozaki. T., Yatzurugi. Y.. and Akiyanla. N. (1970).Concentration and behavior of carbon in semiconductorsilicon. J. EI. Cllelll. Soc., 117. p. 1566.

4. Chipman, J., Alfred. R.M .• GOIt. L.W., Small. R.B ..Wilson. D.M.. Thomson, C.N.. Guernsey. D.L., andFulton. J.c. (1952). The solubility of carbon in molteniron and in iron-silicon and iron-manganese alloys.TrailS ASM, 44. pp. 1215-1231.

5. Novokhatskiy. I.A., and Belov, B.F. (1969).Concentration dependence of the oxygen solubility inmetals. Izl'. Ak"d. Nm,k SSSR Metal/y, 3. pp. 15-24.

6. ShevlSov, V.E. (1979). Solubility of oxygen in high-sili­con iron melts. RII.'.'. Met., 3. pp. 56-58.

7.Shevtsov, V.E., Brovkov, V.A., and Lekhmcls, V.L.(1977). Thermodynamics of solution of oxygen in Fe-Simelts. Steel USSR. 7, I I. pp. 615-617.

8. Shevtsov, V.E. (1979). Thermodynamics of oxygensolutions in high silicon-iron mehs. hv, V. U.zCllenwy" Metall.. II. pp. 5-7.

9. Hirata, H., and Hoshikawa, K. (1990). Oxygen solubili­ty and its temperature dependence in a silicon melt in

THE REFINING OF SILICON AND FERROSILICON

equilibrium with solid silica. J. Cryst. Growth, 106. pp.657-664.

10.JANAF thermochemical tables. (1971 - 1982). NSRDS­NBS 37. with supplements up to 1982.

II. Rein. R.H., and Chipman, J. (1965). Activities in theLiquid solution Si02-CaO-MgO-AI,O, at 1600 dc.Tral/,\'A1ME, 233. pp. 415-425.

12. Elliott, J.F., Gleiser, M., and Ramakrishna, V. (1963).Thermochemistry for steelmaking. Addison-Wesleyrub!. Co. Inc., Pergamon Press.

13. Tuset, J.K. (1985). Principles of silicon refining.Proceedings [men/mianol Semina,. 011 Refilling lIIId

Alloying of Liquid Aluminium and Ferro-Alloys. cd.Engh, T.A., Lyng, S., and Oye H.A. Aluminium-Verlag.Dusseldorf. pp. 50-69.

14.Ageev, Y.A., Minaev. V.M .. Tolstoguzov, N.V ..Pankrashkin, Yu. A., and Arehugov, C.A. (1979). Theequilibrium distribution of aluminium between siliconalloys and synthetic slags. I-:.v. V.U.Z Chemaya Mewll..6. pp. 43-47.

15. Schurmarul, E.. Funders. P.. and Litterscheidt, H.(1975). Dampfdruck des Calciums uber Calcium­Silicium- un Calciull1- Alull1inium-Schmclzcn sowictiber Calcium-Aluminium-Silicium Legicrungen. Arch.Ei,\'enhiiltenwe,\'.. 46. pp. 473-476.

16.0rong. T. (1967). Refining of rerrosilicoll with respectto AI, Ca and Ti (in Norwegian). SINTEF/NTH, Jun.1967.

17. Skei, B. (1990). Relining of silicon alloys by the TinjeclProcess. 48th Electric Furnace Conference, NewOrleans.

18. Grabmaier, F. (1981). Silicol/. Springer-Verlag, Berlin.p. 198.

199